Traffic sign detection based on HOG and PHOG using binary SVM and k-NN

Traffic signs serve important functions on the road. Drivers can easily determine their directions and vehicle speeds by paying attention to traffic signs. However, it is only natural that sometimes drivers misjudge the position and meaning of traffic signs that they ignore them and in the worst case scenario, got involved in accidents. Therefore, technological improvements allow the development of a traffic sign detection system that assists drivers in their commute. Some challenges in this kind of research are weather condition, traffic signs positions, and obstructed view of traffic signs by the other vehicles and advertisement boards. This research has developed a model of traffic sign detection system using color segmentation by transforming the RGB color range into HSI to detect traffic signs based on the colors of red, yellow, blue and green, which are further morphologically processed to minimize searching scope with erosion and labeling. In this method, earlier processes reveal traffic sign candidates that make a Region of Interest (ROI) for its features to be extracted using the Histogram of Oriented Gradient (HOG) or the Pyramid Histogram of Oriented Gradient (PHOG). These are then ensued by a classification process using the Support Vector Machine (SVM) or k-NN to finally determine whether those candidates are traffic signs or not. Results show that this system has traffic sign detection accuracies of 79.03% with PHOG + SVM, 82.01% with HOG + SVM, 81.19% using PHOG + k-NN, and 77.81% using HOG + k-NN.